1.线性回归_红酒数据集_

import numpy as np class LinearRegression: def __init__(self, n_iter=200, eta=1e-3, tol=None): # 训练迭代次数 self.n_iter = n_iter # 学习率 self.eta = eta # 误差变化阈值 self.tol = tol # 模型参数w(训练时初始化) self.w = None def _loss(self, y, y_pred): '''计算损失''' return np.sum((y_pred - y) ** 2) / y.size def _gradient(self, X, y, y_pred): '''计算梯度''' return np.matmul(y_pred - y, X) / y.size def _gradient_descent(self, w, X, y): '''梯度下降算法''' # 若用户指定tol, 则启用早期停止法. if self.tol is not None: loss_old = np.inf # 使用梯度下降至多迭代n_iter次, 更新w. for step_i in range(self.n_iter): # 预测 y_pred = self._predict(X, w) # 计算损失 loss = self._loss(y, y_pred) print('%4i Loss: %s' % (step_i, loss)) # 早期停止法 if self.tol is not None: # 如果损失下降不足阈值, 则终止迭代. if loss_old - loss < self.tol: break loss_old = loss # 计算梯度 grad = self._gradient(X, y, y_pred) # 更新参数w w -= self.eta * grad def _preprocess_data_X(self, X): '''数据预处理''' # 扩展X, 添加x0列并置1. m, n = X.shape X_ = np.empty((m, n + 1)) X_[:, 0] = 1 X_[:, 1:] = X return X_ def train(self, X_train, y_train): '''训练''' # 预处理X_train(添加x0=1) X_train = self._preprocess_data_X(X_train) # 初始化参数向量w _, n = X_train.shape self.w = np.random.random(n) * 0.05 # 执行梯度下降训练w self._gradient_descent(self.w, X_train, y_train) def _predict(self, X, w): '''预测内部接口, 实现函数h(x).''' return np.matmul(X, w) def predict(self, X): '''预测''' X = self._preprocess_data_X(X) return self._predict(X, self.w)